Developing a Global Method for Normalizing Economic Loss from Natural Disasters

The study of past natural catastrophes provides valuable insights into assessing our present-day risk to natural hazards. For example, increases in economic losses from natural catastrophes have been observed globally over the past several decades (Bevere et al. 2019; Duggar et al. 2016; Munich Re 2012; Wallemacq et al. 2018). Determining the factors that contribute to this trend is crucial to extreme event and climate-change research. For present-day risk assessments, catastrophe models—which combine historical event information with current demographic, building, physical, and financial data—are used to model the potential losses caused by catastrophes for specified areas. Thus, our understanding of past events and the impact that they may have today has important implications for natural disaster risk management, mitigation, and adaptation.

Debate remains as to the degree to which increases in economic losses can be attributed to changes in natural processes, exposure, or vulnerability, and whether this trend holds true after normalizing past events to present economic conditions (Bouwer 2018; Handmer et al. 2012; Hoeppe 2016; Pielke 2021). Changes in natural processes include changes in the severity or frequency of natural hazards. Changes in exposure (i.e., physical capital stock exposed to natural catastrophes) are driven by economic factors, such as inflation, population shifts, and economic development. Changes in vulnerability include the adoption of disaster risk reduction or adaptation efforts. To isolate temporal natural and economic factors, past exposure is normalized to present-day conditions by accounting for changes in inflation, population, and economic activity. Normalization often does not account for changes in vulnerability over time. Real reductions in vulnerability are assumed to be modest for many perils and regions (Miller et al. 2009); however, additional research is needed to understand the role of changing vulnerability (Bouwer 2018).

Conventional methods adjust for changes in economic activity using either the value of capital stock or GDP. In choosing between these two measures, most studies elect to use GDP due to its widespread availability. However, capital stock may be preferred because it directly measures the value of physical assets that can be damaged by a natural catastrophe—precisely what normalization approaches seek to control for. In contrast, GDP measures the value of goods and services produced by an economy in 1 year, making it an indirect measure of the value of damageable property.

The purpose of this study was to develop a global approach to normalize past exposure to current levels using the value of capital stock. Capital stock can be estimated for an individual country using the perpetual inventory method and a time-series of gross fixed capital investment. The perpetual inventory method is an accounting method that measures the accumulation of annual gross fixed capital investment while deducting for the value of assets that have reached the end of their service lives. Using capital stock estimates for normalization, we conducted an empirical analysis of normalized hurricane disaster losses for the continental United States (US) from 1930 to 2017. We assessed the robustness of our normalization method by comparing normalized losses to losses derived using publicly available data on capital stock and GDP.

The remainder of the article is structured as follows: First, we review catastrophe loss normalization methods. We present the methodology used to derive capital stock values and the methodology used for the empirical analysis of natural catastrophe losses. Then, results of the empirical analysis are presented and, finally, we discuss the results and implications.

To observe temporal loss trends, past catastrophe losses are “normalized” to present-day values. The conventional method for catastrophe loss normalization is attributed to Roger Pielke, Jr., and coauthors (e.g., Downton et al. 2005; Pielke and Landsea 1998; Pielke et al. 2008, 2003; Vranes and Roger 2009; Weinkle et al. 2018). Using this method, past exposure conditions are adjusted to present values by accounting for changes in inflation, population, and economic activity. Adjusting for inflation accounts for changes in the value of currency over time, while increases in population and economic activity suggest that more people and property are located in exposed areas. To normalize economic losses, the following equation is used:where $c$ = value that current year losses are normalized to; and $t$ = year in which the loss occurred. The gross domestic product (GDP) deflator adjusts for inflation, and the remaining two terms adjust for changes in population in the damaged area and (real) national wealth per capita. Wealth is used to measure economic activity and is adjusted to a per-capita basis to more precisely reflect population changes occurring in the damaged area. In smaller countries or countries with limited data on the damaged-area population, the per-capita adjustment is less necessary (or practical) and is often excluded (Neumayer and Fabian 2011; Pielke et al. 2003). In either case, Eq. (1) produces the estimated loss of a catastrophe as if it occurred in the current year, with current levels of exposure, and assuming constant vulnerability of the structures.

Different macroeconomic data are used to measure economic activity (wealth) using this approach. Commonly, wealth is defined as the total value of tangible (e.g., property, physical capital) and financial (e.g., cash, stocks, bonds) assets. For catastrophe loss normalization, researchers prefer to measure economic activity as the value of capital stock, as it measures the value of damageable property and excludes financial assets. Capital stock is defined as the total value of fixed assets purchased for long-term use, including dwellings, other buildings and structures, machinery and equipment, and cultivated assets such as crops and livestock. Yet, most researchers use GDP as a proxy for capital stock due to limited availability of national capital stock data (Munich Re 2012). Capital stock may be preferred over GDP because it measures physical assets that can be damaged by a natural catastrophe; in contrast, GDP measures the value of all finished goods and services produced in 1 year, and serves as an indirect proxy.

A number of authors have catalogued natural catastrophe loss normalization studies (e.g., Bouwer 2011, 2018; Pielke 2021). Of the identified studies, most normalize catastrophe losses using the conventional method (Eq. 1) with varying measures for and assumptions about economic activity. While some studies use the value of capital stock, most use GDP due to the lack of available data on capital stock values. In recent years, normalization studies continue to utilize the conventional method (e.g., Weinkle et al. 2018), but some build on this approach with modifications. For example, Martinez (2020) argues that a building price deflator is more suitable than the GDP deflator for capturing the price of replacing damaged property. Neumayer and Fabian (2011) developed a method that normalizes catastrophe losses as a dimensionless ratio of actual loss relative to potential loss in a damaged area at the time of the event. Grinsted et al. (2019) build on the work of Neumayer and Fabian (2011) by including an additional term that measures the area of destruction to account for subnational spatial differences in exposure. In both cases, GDP is used as a proxy for wealth because of the availability of GDP information.

There are limitations to using GDP for normalization that necessitate further study. First, the use of GDP rests on the assumption that changes in GDP have (and will continue to be) correlated with changes in capital stock (Munich Re 2012; Neumayer and Fabian 2011). In theory, this assumption may not be valid because GDP measures a 1-year flow of economic exchanges that includes tangible goods and nontangible services. Although it is possible that GDP can capture economic losses from business interruption following a natural catastrophe, it does not measure accumulated capital stock (Neumayer and Fabian 2011). Further, the correlation between GDP and capital stock is not necessarily static, and may carry systematic bias for certain countries and/or time-periods. Although Grinsted et al. (2019) found GDP to be a stable proxy for capital stock in the US, other researchers have estimated that the ratio of the value of national capital stock to output (GDP) can vary between two and four (D’Adda and Scorcu 2003; Neumayer and Fabian 2011) depending on the level of income per capita. For example, low-income countries have a capital–output ratio of about 1.4, middle-income countries have a capital–output ratio of about 2.2, and high-income countries have a capital–output ratio of about 3.1 (King and Levine 1994).

Second, most normalization studies are geographically limited to the US, several countries in Europe, the Caribbean and Latin America, China, and India. Additional research is needed to determine whether the conventional normalization approach is valid for other countries and regions outside of the US (Neumayer and Fabian 2011). One study created a global normalized catastrophe database that covers catastrophe losses in both developed (Australia, Canada, Europe, Japan, South Korea, US) and developing (Caribbean, Central America, China, India, the Philippines) regions (Miller et al. 2009). Losses are trended using inflation, GDP, and national population statistics. The study applied the conventional normalization method, but recognized that results may be imprecise because of difficulties obtaining reliable exchange rate data and the assumption that GDP correlates with capital stock.

Third, the conventional normalization method does not account for temporal changes in vulnerability (Miller et al. 2009). Urbanization and higher concentrations in wealth over time may lead to more investments in natural disaster mitigation and adaptation to protect against natural hazards (Bouwer 2011). Mitigation and adaptation investments may include infrastructure investments (e.g., flood protection) and resilient engineering and design enhancements (e.g., building codes, construction materials), which are likely to vary by region. Population shifts (growth or reductions) may also contribute to a region’s vulnerability. For example, Kühnert et al. (2006) have shown that metropolitan scale affects the per-capita amount of many types of infrastructure. Some normalization methodologies attempt to account for changing vulnerabilities. For example, one study corrects for temporal increases in building resilience to high wind speeds (Crompton and McAneney 2008). Global accounting of vulnerability changes is difficult due to limited and nonstandard data. One study found that rising income (measured by GDP per capita) coincides with a global decline in vulnerability to flood hazard (Jongman et al. 2015). Declines in vulnerability are reflected in decreasing mortality and losses as a share of the people and GDP exposed to inundation.

Additional research is required to assess the validity of normalizing catastrophe losses using GDP and determine whether using capital stock values is preferred. Globally, since GDP figures are more readily available, GDP is a frequently used and commonly accepted proxy for wealth. In some countries, such as the US, GDP may be a valid proxy (Grinsted et al. 2019); however, this may not be the case for all countries or for cross-country comparisons. Rather, capital stock, which is accumulated over decades, may be a more appropriate measure for catastrophe loss normalization (Schmidt et al. 2009).

The purpose of this study is to develop a global method to normalize catastrophe losses to present day using capital stock. Our method builds on the normalization techniques developed by Pielke and coauthors by using capital stock in place of GDP. Capital stock values are derived using a sufficiently long time-series of capital stock investment (a component of GDP) using the perpetual inventory method. The perpetual inventory method is an accounting method that measures the accumulation of annual gross fixed capital investment while deducting for the value of assets that have reached the end of their service lives (OECD 2009). Gross fixed capital investment data are widely available, and the perpetual inventory method is the standard approach use by national statistics offices for estimating capital stocks (United Nations 2009; OECD 2009).

The perpetual inventory method estimates gross fixed capital stock by aggregating annual gross fixed capital investment over time. Gross fixed capital stock is defined as the stock of fixed assets surviving from past investment and revalued at current-year prices (United Nations 2009). Using the perpetual inventory method, fixed assets purchased in the past are revalued to current-year prices using appropriate price indices for fixed assets. In effect, gross fixed capital stock is a measure of the replacement cost of fixed assets, accounting for both inflation and real changes in the value of capital stock.

Gross fixed capital investment data were collected from international and national sources. Outside of the US, gross fixed capital investment data are available from the United Nations (UN) National Accounts Main Aggregates Database. The database, which is maintained by the UN Statistics Division National Accounts Section, provides national accounts tables for more than 200 countries and areas of the world beginning in 1970. Gross fixed capital investment is measured as the total value of fixed assets purchased less disposals in a given year. Fixed assets include dwellings, nonresidential buildings and structures, machinery and equipment (including telecommunications equipment and computer software and hardware), cultivated resources (e.g., livestock, managed crops, and forest), and intellectual property products (e.g., research and development, mineral exploration, literary or artistic originals) (OECD 2009). In the US, gross fixed capital investment data were collected from the US Bureau of Economic Analysis (BEA) Fixed Assets Accounts. The BEA’s Fixed Assets Accounts include statistics on gross fixed capital investment available by asset type and ownership (e.g., private structures, private equipment, and government fixed investment). Gross fixed capital investment data going back to 1929 are available from the BEA.

The perpetual inventory method estimates gross fixed capital stock for a target year by accounting for the accumulation of gross fixed capital investment while deducting for the consumption of assets that have reached the end of their service life. The following equation is used:where $Y$ = target year of the gross capital stock calculation; $y_0$ = first year of the investment data series; $y$ = investment year ($y_0\le y\le Y$); $s$ = capital stock type (e.g., dwellings, nonresidential structures, nonresidential equipment); ${\mathrm{GFCI}}_y^s$ = gross fixed capital investment of stock type $s$ in year $y$; ${\mathrm{RR}}^s$ = retirement rate per year of stock type $s$; PI = capital investment price index; and ${\mathrm{ISL}}^s$ = initial stock level of stock type $s$ in the year prior to $y_0$. Inflation is implicitly measured by the investment price index (PI), which is obtained by dividing the current value of gross fixed capital investment (${\mathrm{GFCI}}_y^s$) by its real counterpart.

The perpetual inventory method accounts for stock-type variations in value, accumulation, and retirement. Therefore, it is necessary to obtain gross fixed capital investment data by stock type. Gross fixed investment data from the UN are provided as a total value. To isolate stock-type variations, we split the aggregate gross fixed capital investment value by the share attributed to each stock type. Capital stock type shares were derived for each country from a variety of sources, including the Organisation for Economic Co-operation and Development (OECD), country-specific national accounts offices, and expert judgment. For OECD countries, capital stock shares were obtained from OECD.Stat, a statistical database for OECD countries and select nonmember countries. For non-OECD countries, stock-type shares were obtained from individual national accounts offices. In countries where capital stock shares were unavailable, stock-type shares were estimated using a combination of peer-country data, statistical analysis, and expert judgment. In the US, the BEA provides fixed capital investment data by stock type, which were used to estimate capital stock using the perpetual inventory method.

Using gross capital stock values derived from the perpetual inventory method for each country, our normalization equation (AHN20 method) is specified as follows:where $c$ = value current year losses are normalized to; $t$ = year in which the loss occurred; and capital stock is of stock type $s$ (residential, nonresidential structures, and nonresidential equipment) in year $t$ or $c$. The population term adjusts for changes in population in the damaged area. Using this approach, losses are normalized using the ratio of the value of current-year capital stock to loss-year capital stock. Inflation is implicitly included in the capital stock values. In the US, we applied the per-capita adjustment to estimate capital stock value in damaged areas. The per-capita adjustment was used as a proxy to account for property growth in damaged areas. Following Weinkle et al. (2018), we used county-level population to measure population growth in damaged areas. An alternative approach would be to apply a housing unit adjustment, which directly captures residential property growth (Weinkle et al. 2018). In some cases, a housing unit adjustment may be preferred; however, this approach does not capture growth in nonresidential structures or equipment in the damaged area. Outside of the US, where damaged-area population data are not always readily available or necessary, the normalization equation simplifies to exclude per-capita adjustments.

To assess the robustness of the AHN20 normalization method, we conducted an empirical analysis of normalized hurricane disaster losses for the continental US from 1930 to 2017. Losses are normalized to 2018, representing the most recent year with available data, and are compared with established US normalization methods. In the US, extensive research has been conducted on normalized hurricane losses. Base damage loss estimates from US hurricanes were obtained from the Supplementary Information provided by Weinkle et al. (2018) (see Supplemental Materials). This hurricane loss dataset has been continuously maintained, with previous versions compiled for 1925–1995 (Pielke and Landsea 1998) and 1900–2005 (Pielke et al. 2008), and is widely used in the insurance industry, research and policy settings. It includes loss estimates from 1900 to 2017 for all landfalling tropical storms with hurricane-force winds ($119\mathrm{km}{\mathrm{h}}^{-1}$) as defined by the National Oceanic and Atmospheric Administration (NOAA). We limited our analysis to 1930–2017 due to the availability of economic data from the BEA, which begins in 1929. Constructing economic data prior to 1929 requires additional data and simplifying assumptions, which was beyond the scope of this study. The loss dataset includes 144 landfalling hurricanes for 1930–2017. From these, we excluded 10 events without base (nominal) loss estimates and eight events with multiple US landfalls. We excluded the eight multiple-landfall events because damage estimates were reported in total and not separated by affected areas in the dataset.

Historical losses were normalized to 2018 values using our normalization method (AHN20) and variations of the conventional normalization method developed by Pielke and coauthors: (1) Pielke and Landsea (2018) (PL18); (2) Weinkle et al. (2018) (CL18); and (3) Pielke and Landsea (2003) (PL03). All three methods follow the same approach (Eq. 1), using different measures for economic activity. Further information for PL18 and CL18 is available from Weinkle et al. (2018). For PL03, further information is available from Pielke et al. (2003). PL18 adjusts historical loss data for inflation, per-capita capital stock, and population changes of damaged counties. CL18 adjusts for inflation, capital stock per housing unit, and housing-unit changes of damaged counties. PL03 adjusts historical loss data for inflation, per-capita GDP, and population changes of damaged counties. Capital stock estimates for the US are available from the BEA and measured using estimates of current-cost net stock of fixed assets and consumer durable goods. Capital stock, GDP, and inflation data were obtained from the BEA. Affected county population data were obtained from Weinkle et al. (2018) (see Supplemental Materials).

Figs. 1(a–d) show normalized hurricane damage losses for the continental US for 1930–2017 using AHN20, PL18, CL18, and PL03 methodologies, respectively. Total normalized losses over the 87-year study period range between USD 1.1 trillion (PL03, PL18) and USD 1.3 trillion (AHN20) across methodologies, with an annual average ranging between USD 12.5 billion (PL03) and USD 15.3 billion (AHN20) per year. The figure shows an 11-year centered moving average, indicating losses on a decadal scale. The AHN20 methodology is consistent with the conventional normalization methods, where losses on a decadal scale were higher in the earlier part of the twentieth century, lower in the 1970s and 1980s, and then higher again in the first decades of the twenty-first century (Weinkle et al. 2018). Fig. 2 compares normalized hurricane damage losses for the continental US derived using the AHN20 and three other normalization methods on log-log scales. The figure further confirms that AHN20 is broadly similar to the other methods, but it also shows that AHN20 produces normalized losses that are (overall) 23% higher than PL18, 16% higher than CL18, and 26% higher than PL03.

We validated our normalization method using hurricane losses for the continental US from 1930 to 2017. Our method, which used gross fixed capital stock, produced normalized loss estimates comparable to the conventional method using publicly available data on net fixed capital stock and GDP. Our method yielded catastrophe losses that were 16%–23% higher than normalization using net capital stock (CL18, PL18) and 26% higher than GDP (PL03). In general, differences in normalized losses were higher in earlier decades and lower in recent decades.

Following Weinkle et al. (2018) and Martinez (2020), it can be useful to compare normalized hurricane damage losses in the US against long-term trends in hurricane landfall frequency and intensity to determine if they share similar statistical properties. We find similar correlations between normalized losses and US hurricane and major hurricane (category 3+) landfalls across normalization methods (Table S1 in the Supplemental Materials). Our findings are consistent with Weinkle et al. (2018) and Martinez (2020). In addition, we find similar agreement in decadal trends across normalization methods (Fig. 1). This coincides with a close relationship between gross capital stock, net capital stock, and GDP growth in the US, as discussed below and shown in Fig. 3.

The difference in normalized losses can be largely attributed to (1) differences in measuring capital stock; and (2) using GDP as a proxy for capital stock. The AHN20 method measured capital stock using the perpetual inventory method. Our measure for capital stock differs from BEA’s current-cost net stock of fixed assets and consumer durable goods (used for PL18 and CL18 normalization) in two ways. First, we measured capital stock as gross capital stock, whereas BEA produces estimates for net capital stock. Second, we limited capital stock to include only insurable assets (residential, nonresidential structures, nonresidential equipment) while excluding consumer durable goods and fixed assets not typically included in the insurance market (e.g., public infrastructure, such as highways and streets, sewers, and water systems). Gross fixed capital stock is defined as the stock of fixed assets surviving from past investment and revalued at current-year prices (United Nations 2009). Gross fixed capital stock differs from net fixed capital stock in that it does not account for depreciation of surviving fixed assets. In effect, gross fixed capital stock measures the “replacement” cost for current fixed assets as if they were purchased new, whereas net fixed capital stock measures the market cost for current fixed assets at their current state of depreciation or obsolescence (OECD 2009). Gross capital stock may overvalue deprecated or partially obsolete fixed assets compared to net stock. Current replacement costs (excluding depreciation) are useful for the purpose of normalization because they may provide an upper bound of catastrophe damage losses.

Previous research argues that the value of capital stock is a better measure for property value than GDP for the purpose of loss normalization (Neumayer and Fabian 2011; Pielke et al. 2003; Schmidt et al. 2009). GDP is used as a proxy for capital stock under the assumption that changes in GDP will be correlated with changes in capital stock. Yet, the ratio of the value of capital stock to GDP can range between two and four for some economies, but will differ across countries (Neumayer and Fabian 2011). Fig. 3 shows indexed growth for US GDP per capita, net fixed capital stock per capita and per housing unit, and gross fixed capital stock per capita from 1929 to 2018. Net fixed capital stock is shown per capita and per housing unit to illustrate the differences between PL18 and CL18 normalization. Fig. 3 suggests that gross stock is broadly similar to other index sources, but that it tends to be greater than net fixed capital stock and GDP. Net stock per housing unit (used in CL18 normalization) has grown more slowly than all per-capita figures due to differences in population and housing growth in the US. In some cases, a housing unit adjustment may be preferred, as it directly captures residential property growth; however, normalization based on housing units may be distorted in areas where commercial property growth is significantly different than housing unit growth (Weinkle et al. 2018). In this study, we chose to normalize using a per-capita adjustment because it has broader applications globally where population data are more readily available than housing unit data. An alternative approach could apply a housing unit adjustment.

The purpose of this study was to develop a global approach to normalizing exposure to current values using capital stock. The use of capital stock may be preferred over GDP for normalization because it directly measures the value of physical assets that can be damaged by a natural catastrophe. For countries other than the US, GDP may be a poor proxy for capital stock. Our normalization method has implications for enhancing catastrophe models using historical events and for analyzing natural catastrophe and climate change loss trends. Our method builds on conventional normalization approaches pioneered by Pielke and coauthors by using capital stock instead of GDP to measure economic activity outside of the US. Capital stock can be estimated for each country using the perpetual inventory method and a sufficiently long time-series of gross fixed capital investment. Gross fixed capital investment data are available for the US from BEA, and for more than 200 countries and areas from the United Nations.

Results from our empirical analysis of normalized catastrophe losses in the US demonstrate that our method is a valid approach for normalizing past exposure to current values. Using capital stock provides results that are broadly similar to established normalization methods presented in this paper. Our results, which utilize gross capital stock, are systematically higher than normalized losses that use net capital stock or GDP as the measure of economic activity. Differences can be attributed to inherent differences between measures of gross and net capital stock and GDP. Our normalization approach offers applications for climate change and natural disaster loss trending. In the US, we find similar agreement in decadal trends across normalization methods using GDP and capital stock where losses on a decadal scale were higher in the earlier part of the twentieth century, lower in the 1970s and 1980s, and then higher again in the first decades of the twenty-first century. This coincides with a close relationship between gross capital stock, net capital stock, and GDP growth in the US. Future research is needed to apply our normalization method to a global loss dataset for the purpose of identifying global loss trends.

Some or all data, models, or code used during the study were provided by a third party. Third-party data include economic data (gross domestic product, gross fixed capital investment, current-cost net stock of fixed assets and consumer durable goods) and damage loss estimates from US hurricanes. Direct requests for these materials may be made to the provider as indicated in the Acknowledgments. Some or all data, models, or code that support the findings of this study are available from the corresponding author upon reasonable request. Normalized loss estimates from US hurricanes are available from the corresponding author.

The manuscript was prepared by Anthony Hanson, Brian Alstadt, and Austin Nijhuis. All authors were responsible for the conception and design of the work. Brian Alstadt and Austin Nijhuis were responsible for data collection and analysis. Austin Nijhuis drafted the article. Anthony Hanson and Brian Alstadt provided critical revision and editing of the article. Economic data used in this study for the United States were obtained by the US Bureau of Economic Analysis. Outside of the US, economic data were obtained from the United Nations and the Organisation for Economic Co-operation and Development. Base damage loss estimates from US hurricanes were obtained from Weinkle et al. (2018) (see Supplemental Materials).